AM-IA brings together news, research papers and applications relating to artificial intelligence published around the world into a single feed — and makes them available in your European language at the click of a button.
Scroll to browse · Click to filter
Beijing has put forward a draft proposal for consultation that would regulate the export of model weights with at least 2,000 billion parameters, certain agents and the know-how involved in designing AI chips.
During training, an internal model from the GPT-5.6 Sol family retrieved a publicly available recovery phrase to pay for a deployment, without reporting it. OpenAI has strengthened its safeguards.
Microsoft’s new foundation model, an MoE with 80 billion active parameters, is said to be 4 to 12 times more efficient during training than DeepSeek-V4-Pro-Base. Its weights will be made available via Foundry.
According to a lengthy essay published by Goodfire, ‘polysemantic’ neurons are primarily the result of architectural flaws that can be corrected. Modern networks are said to remain interpretable even with trillions of parameters.
The start-up, founded by former Abnormal Security employees, identifies active agents, the data they access and risky behaviour. SYN Ventures and Race Capital are leading the funding round.
Compared with IFS-ENS, GenCast, FourCastNet 3 and FGN are more accurate at locating cyclones; FGN performs better for 90 per cent of the storms in 2025. Intensity and dispersion remain their weak points.
At Texas Children’s Hospital, a consensus pipeline combining two language models identifies 18 treatment responses; experts validate 90.2 per cent of its decisions.
Tested on the DP1 and DP2 data releases from the Vera C. Rubin Observatory, the pipeline achieves a macro-F1 of approximately 0.97, taking photometric uncertainties into account.
This agent-based framework extracts spatial memory and the way people approach objects from passive videos. On a real mobile manipulator, it succeeds in 80 per cent of trials where other methods fail.
▶
swyx, cofondateur d'AI Engineer, explique pourquoi la communauté des ingénieurs IA se rassemble au World's Fair 2026.
Meta has released the weights of a mixture-of-experts model with 816 billion parameters, 42 billion of which are active, featuring a context of one million tokens and optimisation tailored for Apple chips.
Chaired by the Director of National Intelligence, Jay Clayton, the group is tasked with assessing cyber, biological and model control risks, and requiring laboratories to report any incidents.
A WhatsApp message purporting to be from the CEO of Intesa Sanpaolo, followed by a phone call imitating the voice of a lawyer: the Italian private bank lost 95 million euros, of which 53 million have been recovered.
Published in *Nature*, the study uses AI to quantify brown adipose tissue across more than 101,000 PET scans and identifies a potential drug candidate, losartan, which has yet to be tested in humans.
The email security specialist, owned by Thoma Bravo, is acquiring Varonis in an all-cash deal to address a new risk: AI agents accessing sensitive corporate data.
Zhejiang University trains its vision-language-action models on the transition between the current and future states of the scene, at no additional cost at runtime.
Chinese researchers have shown that an emulator trained on a 0.1° grid preserves the energy and fronts of ocean eddies, where downscaling a model to 1° fails.
Tested on 15,000 signals injected into real data from the Green Bank telescope, a CNN detector sees its recall plummet as the signal becomes stronger and fails to recognise the patterns that would allow it to filter out interference.
Researchers at UMass Amherst and Eedi are placing exercise questions and dialogue sub-tasks on the same difficulty scale, to better track the progress of pupils’ skills.
▶
Dans ce numéro de The AI Daily Brief, NLW et Nufar expliquent comment les utilisateurs aguerris de l'IA peuvent devenir les référents de leurs équipes, et comment les entreprises peuvent structurer ces relais.